Nvidia's Next-Gen GPU Could be Coming to Intel Foundry

DigiTimes reports Nvidia is considering using Intel Foundry beginning in 2028 for parts of its Feynman GPU family, outsourcing the GPU compute die to TSMC while potentially using Intel 18A (or 14A) for the I/O die and having Intel handle up to 25% of advanced packaging via EMIB. Intel CFO David Zinsner has said advanced-packaging deals are increasingly exceeding $1 billion, and the combination of potential Nvidia and rumored Apple engagements would validate Intel's foundry strategy, expand packaging revenue near-term and help diversify customers away from TSMC; however the report is unconfirmed and contingent on yield improvements.

Analysis

Market structure: A confirmed Nvidia/Intel packaging+I/O die deal (2028, ~25% EMIB share) would immediately validate Intel Foundry economics and create a $1bn+ high-margin revenue stream per large customer; winners = INTC (packaging + nascent foundry), NVDA (supply diversification), AAPL (optional node diversification). Losers = TSM (modest share pressure and pricing leverage loss in specialized packaging) and pure-play packaging vendors if Intel undercuts CoWoS pricing. The supply signal: persistent TSMC capacity tightness is driving demand for second sources and will tighten advanced packaging pricing and lead-times through 2026–2028.

Risk assessment: Tail risks include deal collapse (rumor), Intel yield shortfalls at 18A/14A, or US export policy changes that block flows to key customers — any of which could wipe out >30% of the speculative valuation uplift priced into INTC. Immediate (days) volatility will be rumor-driven; short-term (months) depends on hiring/yield milestones; long-term (2026–2029) on sustained multi-customer traction. Hidden dependency: Nvidia keeping compute die at TSMC limits Intel’s margin capture and makes Intel’s ultimate foundry scale dependent on convincing multiple tier-1 customers.

Trade implications: Favor asymmetric exposure to INTC via limited-risk options and small directional equity sizing: 1–3% portfolio initial longs with staged add-ons on confirmed QoQ hiring/yield milestones; use INTC 12–18 month call spreads to limit downside. Relative-value: pair trade long INTC vs short TSM (smaller size) to play share-shift while hedging semiconductor cyclicity. Volatility: buy INTC calendar or diagonal calls around quarterly releases and sell short-dated calls after positive confirmations; avoid naked short on TSM given moat.

Contrarian angles: Consensus underestimates execution risk — Intel must convert packaging wins into wafer manufacturing to justify capex; history (Intel’s node delays) suggests multiple false-starts are possible. The market may be over-rotating into long INTC on headline wins while underweighting NVDA’s continued TSMC dependence (compute die bottleneck persists). Unintended consequence: more US-based capacity could strengthen USD, pressuring EM-listed fabs and lifting capex-intensive industrial suppliers rather than pure-node leaders.

Key Decisions for Investors

• Establish a 2% long position in INTC equity staged over 4 weeks; if INTC confirms a customer packaging/manufacturing win or posts sustained yield improvement by Q4 2026, add another 2% (target hold 12–36 months). Set initial stop-loss at 20% and take-profit trims at +30% and +50%.
• Buy a limited-risk INTC 12–18 month call spread (e.g., 20–35% OTM depending on current spot) sized to 1–2% of portfolio to capture upside while capping premium; roll or close on missed yield milestones through 2026.
• Implement a pair trade: long INTC (2% exposure) vs short TSM (1% exposure) to express share-shift risk; tighten the short if TSM reports incremental capacity expansion or if NVDA/TMSC confirmations emerge. Rebalance if spread moves >25% in either direction.
• Take a tactical NVDA options calendar spread (3–6 month front-short, 12 month long) sized to 0.5–1% portfolio to hedge supply-diversification headlines without paying elevated near-term IV; exit on Nvidia confirmation of Intel usage or if implied vol drops >30%.